Intel cancels Larrabee GPGPU-chip

Intel has spent several years developing the Larrabee GPGPU-chip that was initially set for release in 2008. The multifunctional chip based on the x86 instruction set was to serve both as a parallel computing processor as well as a graphics card. Earlier this year, it already became known that the performance of the chip as GPU are to be compared to nVidia's GTX 285 card. During the last IDF, two demonstrations of the Larrabee chip were also given.

The last launch date of the chip was officially planned for the first quarter of 2010, but several rumors already mentioned the permanent cancellation of the Larrabee chip. Intel has since officially confirmed this. Though the graphics card is cancelled, the Larrabee project will remain in existence. It will continue the development of software and the SDK package.


Though the exact reason for cancellation remains unknown, it's likely that Intel was falling behind schedule with the development of the chip and matching software. If Intel were to continue the launch of the Larrabee, the card was to compete with nVidia's Fermi GTX 300 and AMD's Radeon HD5000 series, both significantly faster than the GTX 285 that the Larrabee was to be on par with.

The disappointing performance of the chip is not just caused by the delay of the chip itself, but also by the problematic development of the matching software. Though Intel has a lot of experience when it comes to developing x86 software to program the chip, the company isn't very used to graphics drivers. The multifunctional chip was also to retain backwards-compatibility with older drivers such as DirectX 9.0, while also being prepared for the future DirectX 12 version, making drivers difficult to program.

Intel initially planned to release the Larrabee as one of the fastest graphics cards at launch that could also be used for GPGPU applications. Since this won't be achieved in 2010, the graphics card was cancelled altogether. Though the first Larrabee chip will not see the daylight, the project will still continue, making it possible that Intel will release a more powreful Larrabee graphics card at a later point in time

Intel's 32nm Core i3/i5 in pre-order

Over a month before Intel will officially launch its new Core i3/5 dualcore processors, a German webshop has already added the new models as pre-orders. The online catalog shows two Core i3 models, the 530 and 540. These two models feature a clock frequency of respectively 2,93 and 3,06 GHz, which is locked due to the absence of Turbo boost. The CPUs do support Hyperthreading, allowing an operating system to work with four virtual cores. The integrated DirectX 10 GPU in these Core i3 chips is clocked at 733 MHz.

At the moment, these entry-level 32nm dualcores with integrated videochip cost 103,90 and 120,90 for respectively the i3 530 and i3 540. In the Core i5 series, the existing i5 750 quadcore will be joined by four 32nm dualcore 'Clarkdale' chips. These new models are branded Core i5 650, 660, 661 and 670, clocked at 3,2 GHz for the cheapest and 3,46 GHz for the most expensive model. Apart from these higher clock frequencies, the Core i5 models of Intel's new 32nm dualcores also support TUrbo Boost, allowing the frequencies to be increased when adequate cooling is present.

A remarkable processor is the Core i5 661: it is the only chip in the entire Clarkdale family to feature a higher GPU clock frequency, 900 MHz rather than the conventional 733 MHz. This does affect the TDP, which rises to 83 Watt. All other chips have an estimated TDP of 73 Watt. The pricetags of the Core i5 6xx models range from 160,90 for the 3,2 GHz model to 252,90 for the 3,46 GHz chip. The Core i5 660 and 661 can both be pre-ordered at a price of 175,90.

HD5950 to be released in early 2010?

HKEPC reports that AMD is working on a second dual-GPU graphics card in its DirectX 11 line-up. This new model is to be positioned below the HD5970 and will also be equipped with two Cypress chips. The difference is that each core will not feature 1600 but only 1440 active shader units, making it look more like a double HD5850 solution. The clock frequencies of this HD5950 card will be lower than those of the HD5970, otherwise the performance gap between the two would be too small.

Heat production can also be an issue when the original HD5850 clock frequencies are retained; the high-end HD5970 already uses specially selected GPUs which will run at a lower voltage. The Radeon HD5950 is to be equipped with two times 1GB of GDDR5 memory. It's still unclear how fast the HD5950 will be in comparison to an HD5970 or Crossfire HD5850 configuration, given the uncertainties in regards to the final clock frequencies.

Because AMD is already troubled by the huge demand, it seems unlikely that a second dual-GPU card will be released soon. HKEPC expects that the new model will new be introduced until early 2010.

Intel demonstrates processor with 48 cores

Chip manufacter Intel has recently demonstrated a new processor consisting of 48 seperate cores. Intel calls this chip the Single-chip Cloud Computer (SCC). This processor is derived from the Tera design, which was introduced in 2006. Initially, processors based on the Tera architecture were to feature 80 cores, though these have never been produced for industrial applications. The Tera design was a so-called 'proof-of-concept' design. This resulted in the creation of the Single-chip Cloud Computer, codenamed Bangalore.

While the Tera architecture consisted of 80 'simple' floating point processors, the SCC features 48 x86 chips, created using a 45nm process. The surface is specified at 567 mm². The 48 cores are divided in 24 ' tiles'. Intel calls these dual-core tiles. Each of these tiles is capable of running at a different clock frequency to limit power consumption. These tiles communicate with one another by means of a 6x4 2D mesh network, featuring a total bandwidth of 256 GB/s.

The design also integrates four DDR3 memory controllers, capable of adressing up to 64 GB of memory. Despite the specifications, the chip uses no more power than a modern quad-core processor. The usage of this SCC ranges from 25 to 125W, dependant on the required performance.

The name Single-chip Cloud Computing can be explained because the design of the Bangalore is similar to the method datacenters are used for cloud computping. Everything is connected to eachother to gather more computing power. Intel claims that the new chip will also reduce the space required for servers, as well as significantly drop their power consumption.

It's still unclear when Intel will release its new SCC processor. The company does report that it won't take much longer, given the fact presentations have already been given. An SCC chip already rendered a JavaScript-based 3D model. The motherboard used in combination with this chip is branded Copper Ridge and sports an LGA 1248 socket.